Risk Level Prediction of Diabetic Retinopathy based on Retinal Images using Deep Learning Algorithm

Abstract

Diabetic Retinopathy (DR) is an eye illness related with constant diabetes. Diabetes has turned into a worldwide pandemic, with 365 million individuals anticipated to be impacted by 2025. Subsequently, diabetic retinopathy is the most well-known reason for visual deficiency in the United States. The present world is industrialized, and medicines might be compelling, if we get the disease in early stage. Vision is protected, and it is impressively decreased to cripple visual impairment. We are creating and assessing a strategy as a feature of this study for modern retinopathy findings. This work incorporates arranging Diabetic Retinopathy (DR) into two classes (DR, No-DR) and five unique classes (Mild, Moderate, No-DR, Severe, Proliferate) too. Additionally, Data Collection, Data Analyzing, Data Visualization and Data Pre-handling of the gathered information is finished. This work presents an outline of a deep learning algorithmic methodology and gives execution results to a dataset of 5000 Fundus images. While listing the sickness into two unique classes, a total exhibition rate of 89% correct findings is accomplished utilizing VGG-19 deep learning model which is more precise than VGG-16 and CNN model predictions and while characterizing the infection into five unique classes a total exhibition of 90% correct determination is accomplished utilizing VGG-16 deep learning model which is more accurate than VGG-19 and CNN model calculations.